InpressinElectromagneticMedicineandBiology,2009 REVIEW THEEFFECTSOFMOBILE PHONEELECTROMAGNETICFIELDS ONBRAINELECTRICALACTIVITY: ACRITICALANALYSISOFTHELITERATURE AndrewA.Marino,Ph.D. Professor DepartmentofOrthopaedicSurgery LSUHealthSciencesCenter P.O.Box33932 Shreveport,LA71130 3932 Phone:318 675 6180 Fax:318 675 6186 e mail:amarino@lsuhsc.edu SimonaCarrubba,Ph.D. Post DoctoralFellow DepartmentofOrthopaedicSurgery LSUHealthSciencesCenter 1
Abstract We analyzed the reports in which human brain electrical activity was comparedbetweenthepresenceandabsenceofradio frequencyandlow frequency electromagnetic fields (EMFs) from mobile phones, or between pre and postexposure to the EMFs. Of 55 reports, 37 claimed and 18 denied an EMF induced effectoneitherthebaselineelectroencephalogram(eeg),oroncognitiveprocessing ofvisualorauditorystimuliasreflectedinchangesinevent relatedpotentials.the positive reports did not adequately consider the family wise error rate, the presence ofspike artifactsin theeeg, orthe confoundingroleof thetwo different EMFs.Thenegativereportscontainedneitherpositivecontrolsnorpoweranalyses. Almost all reports were based on the incorrect assumption that the brain was in equilibrium with its surroundings. Overall, the doubt regarding the existence of reproduciblemobile phoneemfsonbrainactivitycreatedbythereportsappeared to legitimate the knowledge claims of the mobile phone industry. However it funded, partly or wholly, at least 87% of the reports. From an analysis of their cognitive framework, the common use of disclaimers, the absence of information concerning conflicts of interest, and the industry s donations to the principal EMF journal, we inferred that the doubt was manufactured by the industry. The crucial scientific question of the pathophysiology of mobile phone EMFs as reflected in measurements of brain electrical activity remains unanswered, and essentially unaddressed. 2
Contents 1. Introduction... 4 2. Methods... 5 2.1. Databases... 5 2.2. TheIndependentVariable... 6 2.3. DependentVariables... 6 3. GSMEMFsandBrainElectricalActivity... 8 3.1. EffectsDuringExposure... 8 3.2. Summary... 19 3.3. EffectsAfterExposure... 21 3.4. Summary... 24 4. Discussion... 25 4.1. AnalysisofReports... 25 4.2. ManufacturedDoubt... 30 4.3. Bartleby... 35 5. Acknowledgements... 36 References... 37 3
1.Introduction The seminal question regarding the pathophysiology of the radio frequency and low frequencyelectromagneticfields(emfs)producedbymobilephonesiswhetheroneor both of the fields cause or contribute to the onset of disease. Both EMFs enter the user s brainduringnormalphoneuse;consequentlytheireffectsonbrainmetabolismhavebeen studied intensively. Our purpose was to review the reports in which the measured endpoints directly involved brain electrical activity, and to assess whether the reports providedcrediblescientificevidenceofacause effectrelationship. In the next section we described how the pertinent reports were identified, gave reasons for our treatment of the independent variable, and briefly described the main methods for characterizing brain electrical activity. In many reports the mobile phone EMFswerediscussedinrelationtotheconceptofdose;howeverthereportswerereadily comprehensiblewithouttheneedtodiscussthatclusterofarguments. In Section 3 we inquired critically into the reports that concluded mobile phone EMFs had affected human brain electrical activity, either during or after EMF exposure. Eventhough67%ofthereportswereself reportedaspositive,significantdoubtremained whethermobile phoneemfsaffectedbrainelectricalactivity. In the last section we analyzed the studies globally and explained the basis of our viewthatthedoubtwasmanufacturedbythemobile phoneindustry. 2.Methods 2.1.Databases We searched electronic databases (PubMed, Science Citation Index) using combinationsofanelectricalterm(field,electromagnetic,electric,magnetic,asexamples), a device (mobile phone, cellular phone), and an outcome (electroencephalogram, eventrelatedpotentials,evokedpotentials,brainelectricalactivity)toidentifyenglish language reports that involved the effects of mobile phone EMFs on the brain electrical activity of human subjects. The inclusion criteria were: (1)a reasonable description of the experimental conditions; (2)use of a control group; (3)blinding of the experimental subjectstothetreatment;(4)statisticalevaluationofthedata.theexclusioncriteriawere: (1)the use of thermal EMFs; (2)experiments where the measured endpoints did not includeatleastoneelectrophysiologicaldependentvariable;(3)experimentsthatdidnot include EMF frequencies in the radio frequency range associated with mobile phone technology. All other factors including blinding of the investigators, counter balancing of experimental conditions, performance of sham studies, inclusion of positive controls, corrections for multiple comparisons, the role of artifacts, and the reasonableness of the experimentaldesignwereconsideredwithregardtotheweightgiventothereportrather than to its admissibility as evidence of the ability of the EMFs to affect brain activity. Reportslimitedtolow frequencyemfswerereviewedelsewhere[1]. 4
2.2.TheIndependentVariable GSM technology, the present standard in the mobile phone industry, results in the production of two distinct EMFs. The radio frequency GSM EMF is a high frequency (1 2GHz) pulse lasting 577µs, transmitted every 4615µs (217 Hz burst rate) at 1 2watts, delivered from an antenna located near the head [2]. The low frequency GSM EMF is a 217-Hz magnetic field, about 300mG more or less, produced by the battery current that facilitatestheburstingbehavior(notbytheantenna)[3 5]. Information in the reports regarding both EMFs was generally sketchy, but its absence was not a serious problem.knowing that the subjects were exposed to actual or simulatedgsmemfswassufficientbecauseourfocuswasonthecausalassociationofgsm EMFs and the occurrence of changes in brain electrical activity. Many investigators provided calculations and measurements of the specific absorption rate and/or engineeringdetailsofspecificmobilephones.wedidnotlisttheinformationbecausethe investigatorsdidnotdeterministicallyrelateit(empiricallyortheoretically)totheeffects theyclaimed. 2.3.DependentVariables Brain electrical activity was represented as voltage time series measured from standardized locations (10 20 system) using metallic electrodes glued to the scalp (electroencephalogram(eeg)).inmostexperiments,theeegwasrecordedintheabsence of a specific sensory stimulus (baseline EEG) and compared between the presence and absenceoftheemf,orbetweenpre andpost exposure(fig.1).spectralanalysiswasthe common method for characterizing the baseline EEG; the signals were decomposed by Fourieranalysisintotheircomponentfrequencies,eachrepresentedbyacoefficient(called power, and expressed in units of µv 2 ). The spectral frequencies were band limited, sampled,subdivided,combined,andnormalizedinnumerousways;itwasunnecessaryto listthedetails(whichwererarelythesameindifferentstudies).thealphaband(8 12Hz) wasthefocusinmanystudies. In other experiments brain electrical activity was studied while the subject was respondingtoaspecificstimulusbyexhibitinganevent relatedpotential(achangeinthe EEGduetospecificsensoryorcognitivestimuli[6]).Presentationofasensorystimulusas a relatively rare event(for example, 500-Hz tones among more common1000-hz tones) was a typical technique for studying cognitive processing (oddball paradigm). When a stimulus was presented in the absence of any expected cognitive or behavioral response, wereferredtotheevent relatedpotentialasanevokedpotential[6].asinthebaselineeeg studies,event relatedpotentialswerecomparedbetweenpresenceandabsenceoftheemf or between pre and post exposure. The amplitudes of their various components and the correspondinglatencies(fromthetimeofapplicationofthestimulus)weredeterminedby time averaging the EEG over repeated applications of the stimulus, resulting in 3 6 variables,dependingonthestudy. 5
Figure 1. Experimental procedures used to study the effect of mobile phone electro magnetic fields on brain electrical activity. a)the electroencephalogram (EEG) was recorded in the presence and absence of mobile phone EMFs. In experiments where actual mobile phones were used, the EMFs consisted of radiofrequencyfieldspro ducedbytheantenna, and low frequency magnetic fields produced by the battery current. In the reports where simulated mobile phone EMFs were applied, the subjects were exposed to only radio frequency EMFs. Brain electrical activity was characterized by measuring either baseline EEG or event related potentials superimposed on the baseline EEG as a result of cognitive processing of visual or auditory stimuli. b)brain electrical activity was recorded beforeandafterexposuretomobile phone EMFs,butnotduringexposure. Several more unusual variables were also measured including slow potentials, desynchronizationreactions,eegcorrelation,andmagnetic(asopposedtoelectric)fields. Detailsofthesemethodsaregivenelsewhere[6]. 3. GSMEMFsandBrainElectricalActivity 3.1. EffectsDuringExposure TheinfluenceofEMFsfromamobilephoneontherestingEEGwasexaminedin36 subjectsexposedfor15min,andanincreaseinrelativealphapowerduringandafterfield exposure was described [7]. However, multiple tests were performed (16 electrodes, 6 frequency bands, 2 experimental conditions, and 3 exposure durations were evaluated) withnocorrectionsforfamily wiseerror.anotherlimitationinvolvedtheinteractionofthe EMFswiththescalpelectrodes.Thelow frequencymodulationoftheradio frequencyfield and the low frequency modulation of the battery current both probably produced spike artifacts attheinputoftheeegamplifier[8];thepotentialroleoftheseartifactswasnot adequatelyaddressed(seebelow). In an experiment involving the effect of mobile phone EMFs (900MHz, 217Hz modulation,50µw/cm 2 )ontheeeg(recordedfromc3andc4)in34subjects,noeffecton spectral power was found during a 3.5 min exposure [9]. Positive controls were not 6
included; consequently the sensitivity of the experiment for detecting an effect remained unaddressed. Hietanenetal.[10]studiedtheeffectof5differentmobile phoneemfsontheeeg from 21 derivations in 19 subjects. Ninety t tests were performed (4 brain regions x 4 frequencybandsx5phones)andonlyonesignificantdifferencefromthecontroleegswas found, clearly supporting the investigators conclusion that there was no evidence of an effectofthemobile phoneemfsontheeeg. Croft et al. [11] measured baseline EEG and auditory evoked potentials from 24 subjects in the presence and absence of a mobile phone EMF, and analyzed the data in terms of numerous dependent variables. As was usually the case when many statistical testswereperformed,afewwerepair wisesignificant;theyincludedadecreaseat1 4Hz andanincreaseat8 12HzintherestingEEG,andalteredsound induceddecrementsat4 8Hz, 12 30Hz, and 30 45-Hz in the evoked potential measurements. The investigators concluded that mobile phone EMFs affected neural function, but a more parsimonious explanationwasthatthefewsignificanteffectstheyobservedoccurredbychance. D Costaetal.[12]comparedtheEEGinthepresenceandabsenceofamobile phone EMF. The data was obtained during a series of 5 min on/off cycles and analyzed using contralateral electrode derivations as respective references (for example, O1 derivation referencedtoo2).in12tests(3datasetsx4frequencybands)threeeffects(oneinalpha and two in beta) were found. Considered as a prospective study, the investigators conclusion( TheresultsofthisstudylendsupporttoEEGeffectsofmobilephones )seems warranted because the family wise error rate (PFW) for 3 tests at a pair wise level of p<0.05 is PFW<0.02. This interpretationwas strengthened by their results from a sham analysis (phone in the standby mode) in which 0/12 tests were significant. The investigatorsdidnotexplainthereasonfortheirchoiceofelectrodeconfiguration,which correspondedtothehypothesisthatmobile phoneemfsaffectedbrain electrodeactivity moreatonelocationthananother(ornot),ratherthantheirprofessedhypothesis,which was that they affected brain electrical activity (or not). A limitation in the study was the unaddressed problem of electrode artifacts, which are EMF electrode interactions that havenophysiologicalsignificance. In another experiment involving a typical mobile phone EMF (902MHz, burst frequency 217Hz, maximum power 2W, average power 0.25W), Curcio, et al. [13] describedincreasedspectralpowerat9hzand10hzduringexposure(andaweakeraftereffect).however,theconclusionwasbasedonfive3 wayanovas,onlyoneofwhichwas statistically significant. Moreover, the effect was small and the variance was large. For example, at 9Hz the EEG power while the field was present was (mean ± SD) 0.8±0.3, comparedwith0.7±0.4forthecontrol;theresultat10hzwassimilarlyproblematical. Croft, et al. [14] repeated the study [13] using many more subjects and found increased alphapowerduringemfexposure, usingone tailedtests.amajorshortcoming in the study was the presence of audio noise from the phone circuitry. The investigators 7
reportedthat2subjectshadbeenunabletohearthesound;however,whetheritcouldbe discernedbyanyoftheother107subjectswasnotevaluated. Kleinlogel,etal.[15,16]foundnoeffectofmobile phoneemfsonbrainelectrical activity, but the studies had toomany independent variables(3 different mobile phones), toomanydependentvariables(baselineeegvisualevokedpotentialsandauditoryevoked potentials),andtoofewsubjects(n=15)tohaveanyreasonablechanceofavoidingatypeii statisticalerror. A German group investigated the effects of mobile phone EMFs on brain electrical activity during sleep [17 19]. In the first study [17], the EEG from one derivation was analyzedin12subjectsonsuccessivenightsduringthepresence(8hrs)andabsenceofa mobile phone EMF. Spectral power during REM sleep was increased 5% during EMF exposure but not in the other sleep stages. A finding of 1pair wise significant result in 5 tests(5sleepstagesconsidered)correspondstoafamily wiseerrorrateof0.23,thusthe studyprovidesnoreliableevidencethatemfsaffectedtheeeg.inasecond[18]andthird [19] study (24 and 20 subjects, respectively) the investigators reported failure to demonstrateaneffectofmobile phoneemfsontheeeg. ASwissgroupcontinuouslyrecordedtheEEGof24subjectsduringan8 hournighttimesleepsessionwhilethesubjectswereintermittently(15minon/15minoff)exposed tomobile phoneemfs[20,21];thesubjectsweresham exposedoneweekearlierorlater. Powerspectrawerecomputedandcomparedinfrequencybinsof0.25Hzintherange0 25Hz(100statisticaltests),andtheresultswerepresentedintwopublications.Whenthe investigatorsfocusedonthetimebetweensleeponsetandthefirstepisodeofremsleep, they found many pair wise significant comparisons in alpha (Fig. 2a) [20]. When the choseneeganalysisintervalwasthefirst30minafterlightsoff(theperiodduringwhich the subjects were falling asleep) [21], there were far fewer pair wise significant comparisons (Fig. 2b). The major limitation regarding the investigators conclusion that theyhaddemonstratedanemfeffectontheeegwasthattheypresentedresultsforonly two time intervals. The investigators disclosed neither the number of intervals analyzed northereasonthedatawasnotdividedaccordingtosleepstages,whichisnormallyhow effectsonsleeparestudied.aswascommoninthereportswereviewed,theinvestigators extensivelydiscussedthespecificabsorptionratebutfailedtorelatethediscussiontotheir results.theyalsodidnotdiscussthepotentialartifactsduetoelectrodepick up. Fritzer et al. [22] investigated whether night long EMF exposure affected the spectral power in ten subjects using a separate control group, but found no field induced changesinpoweratanyfrequencyineitherremornon REMsleep.Thefailuretousethe exposedsubjectsastheirowncontrolswasaseriouserror. Lebedeva, et al. [23] recorded the EEG from 16 derivations in 24 subjects, and reported that the average fractal correlation dimension increased during and after exposuretoamobile phoneemf.similarresultswerefoundwhentheemfwasappliedto subjectsduringsleep[24].thestudieswererareexamplesofanexperimentaldesignthat 8
consideredthepossibilitiesthat(1)thestimulus responserelationshipmightbenonlinear, and(2)the subjects might respond differentlyfrom one another. The difficulty with both studies was that fractal analysis of biological time series data is inherently problematical becausethemathematicaltechniquespresumenoise freestationarydata,whichisfarfrom the case for the EEG. The validity of the fractal analysis of the EEG must be established usingsurrogateanalysis[25]orsomeothercontrolprocedure,whichtheinvestigatorsdid notdo. Figure2.Theroleof theanalysisintervalinthedecision regarding whether mobile phone EMFs affected brain spectral power. a) EEG data analyzed for the period between sleep onset and initial REM episode. b) Same data analyzed for the first 30 min period after lights off. The ordinate values are the means (±SE) of the spectral power relative to the control (taken as 100%). The bars indicate the frequencies for which the statistical comparison was pair wise significant at p<0.05. Data in a) and b) from Borbély et al. (20) and Huber et al. (21), respectively. Estonianinvestigatorspublishedfivereportsdealingwiththeeffectofexposureto pulse modulated450-mhzfieldsonbrainelectricalactivity[26 30].Initiallytheyexamined the relative spectral power in the EEG from 23 subjects who each received 10 cycles of 1-minexposurefollowedbya1-mincontrolperiod[30].Onaveragetherewerenoeffects on baseline EEG; however there were also noeffects when the subjects were exposed to light,indicatingthattheexperimentwaswoefullyinsensitive.theinvestigatorsclaimedto 9
haveobservedemfeffectsonthetafrequencieswhenthedatawasevaluatedseparatelyin eachsubject.however,thepost hocanalysiswasunprotectedagainstanexplanationbased on chance. In a subsequent similar study involving 13 subjects (but no positive control) increasedrelativepower(about6%)duetoemfswasreported[29],butinthecontextof numerous3 wayanovas,thegreatmajorityofwhichwerestatisticallyinsignificant. IntwootherstudiesanovelformofnonlinearanalysisoftheEEGtimeserieswas employed,and10of38subjectswereshowntohaverespondedtotheemfbyexhibiting altered brain electrical activity[26, 27]. Unlike multifractal analysis to which the method hassomesimilarities,theinvestigators methoddidnotattempttocharacterizethesource (the brain of the subject); the validity of the method therefore did not depend on the stationarityofthedata.themajorlimitationwasthatonlytenexposedandcontrolepochs were used in the statistical comparisons for each subject, possibly accounting for thelow responserate(10/38).thepossibleinfluenceofspikeartifactswasnotconsidered. TheinvestigatorsreanalyzedtheEEGsobtainedintheirfourstudies(4maleswere deletedfromtheir2005study)toevaluatetheeffectofmobile phoneemfsusingalinear method (spectral power), but without averaging across subjects [28]; for unexplained reasonsonlytheparietalderivationswereconsidered.theyfound13 31%ofthesubjects detected the field, depending on the modulation frequency. The investigators did not explain why they used a linear method after having achieved some success using a nonlinearapproach. AFrenchgroupinvestigatedtheeffectsofGSMEMFsonauditoryevokedresponses from normal and epileptic subjects; the EEG was recorded from 32 electrodes in the presence and absence of the fields and analyzed three different ways in separate publications[31,32,33].first,spectralcorrelationcoefficients,latency,andamplitudeof the time averaged signals were computed from 14 of the electrodes. The reported fieldinduced effects, averaged over the subjects, were decreased N100 amplitude and spectral correlationcoefficientsforbothgroups,areductioninn100latencyinthenormalsubjects butanincreaseintheepileptics,andincreasedp200amplitudeinthenormalsubjects[31]. Then the investigators computed the temporal and spectral correlation coefficients for some individual subjects, using 13 electrodes [32]. Field exposure altered temporal correlation in 7 of 9 normal subjects (increase in 5, decrease in 2) and in 5 of the 6 epileptics (increase in 2, decrease in 3). Frequency correlation was altered in 8 normal subjects(increasein5,decreasein3)andin5epileptics(increasein4,decreasein1).in theirthirdanalysis,thevariableswerereanalyzedandaveragedacrossallelectrodesand allsubjects,andthepreviouslyreportedresultswereconfirmed[33]. Thelargenumberofvariables,andtheirarbitrariness,obviatedthepossibilitythat the conclusions reached by the investigators were reliable. Their strategy of averaging across subjects [29] after the results of a prior analysis showed that different subjects respondeddifferentlytothefields[28]wasparticularlyinexplicable. 10
Bak,etal.[34]evaluatedtheeffectsof450,935,and1800MHzmobile phoneemfs onauditorybrain stemevokedresponsesmeasuredduringandafteremfexposure.each subjectwasexposedfor20minutes,andtheaveragedresponseswerecomparedacross15 subjectsusingmultiplemultiple factoranovas.noeffectsonlatencywerefound. The investigators avowed goal was to shed light on the problem of why earlier studies of the effects of mobile phone EMFs on cognitive processing were contradictory. However,brain stemauditorypotentialsmeasureonlytherateofneuralconductionalong theauditorynervebetweentheearandtheauditorybrain stem;theyarenotameasureof cognitive processing. The investigators recognized this limitation ( It is noteworthy that theauditorybrainstemresponsereflectsthefunctionofonlyaminorpartoftheauditory pathway ),butdidnotreconciletheclearmismatchbetweentheirgoalsandtheirmethods. Further,althoughtheydidnotevaluatethesensitivityoftheirexperiment,thevariancein thedatasuggestedthattheinvestigatorswouldnothavedetectedevenclinicallysignificant brain stempathology.thisreportexemplifiedthemeaninglessnessofnegativestudiesthat alsofailtoincludepositivecontrolsorapoweranalysis. Eulitz, et al. [35] used an oddball paradigm to measure auditory event related potentialsinthepresenceandabsenceofmobile phoneemfs,butfoundnoeffects.when they reanalyzed the data using a time dependent form of spectral analysis in connection with multiple three way ANOVAs, one test (18.75 31.25Hz from the left hemisphere for one of the oddball conditions) was significant. Their conclusion (that the EMFs alter distinct aspects of the brain s electrical response to acoustic stimuli ) was unwarranted because no corrections were made to account for the multiple statistical tests that were performed. German investigators [36] studied the effects of mobile phone EMFs on manifestationsofbrainelectricalactivitythatprecededvolitionalmusclemovement(slow potentials); data was obtained from 30 electrodes. The investigators performed 3 way repeated measures ANOVAs; the factors were EMF(on/off), hemisphere(left/right), and brainregion(threeregions).variousdifferencesforthefactorsandtheirinteractionswere described, and interpreted to mean that there may have been an EMF effect. In two additional experiments [37], the investigators described bilateral decreases in average slowpotentialsfromaparticularcollectionofderivations(fig.4).itseemsimplausiblefor suchspecificeffectstohaveoccurredtwicebychance.ontheotherhand,theinvestigators did not indicate whether the number and identity of the electrodes in each region were chosenbeforeorafterthedatawascollected.anotherprobleminvolvedthechoiceofthe EEGanalysisinterval.Theslowpotentialswereaveragedoverthe500msprecedingmotor activity,eventhoughnumerousotheraveragingintervalswerepossible;noexplanationfor that choice was provided. In a previous study not involving EMFs[38], the investigators usedfourdifferentaveragingintervalsnoneofwhichwerethesameasthoseusedintheir EMFexperiments[37]. 11
Figure4.Meanslowpotential(timeinterval 500to 0 ms from moment of key press) in the visual monitoring task in two independent experiments (from Freude et al. [37]). The lines indicate the electrodederivationsthatwerecombinedtoproduce the indicated bar graphs. The data from the other derivationswasnotusedintheanalysis. Hamblin et al. [39] investigated the effects of EMFs on auditory event related potentials in 12 subjects performing an oddball task who were exposed (30 min) and sham exposed(1weekapart),withthephonemountedontherightside.peakamplitudes and latencies were extracted for target and non target stimuli and averaged over 6 head regions. Three way ANOVA indicated that EMF exposure was associated with a right side reductioninamplitudeandlatencyofthen100wavefornon targetstimuliandanincrease inp300latencyfortargetstimuli. In a second study [40], the specific hypotheses of a reduction of amplitude and latencyofn100tonon targetstimuliandanincreaseinp300latencytotargetduetoemf exposureweretested.nosignificanteffectswerefound,andtheauthorsconcluded the presentresultsdetractfromthepreviouspositiveevidenceandweconcludethatthereis currently no clear evidence in support of a mobile phone related EMF effect on eventrelated potentials. This was one of the few instances where investigators sought to prospectively test a hypothesis gleaned on the basis of a data mining approachin a prior study.theyopinedregardingthereasonsfortheirfailuretosupporttheirhypothesis,but they did not acknowledge that the most probable reason was that the effects of EMFs on brain electrical activity are nonlinearly related to the field, whereas they used linear analysis(anovas). Jechetal.[41]investigatedtheeffectsofEMFsonevent relatedpotentialsrecorded during EMF exposure, and on baseline EEG recorded after exposure. The subjects, who 12
werenarcoleptics,wereeitherexposedorsham exposedonsuccessivedays.thelatencies andamplitudesofvariouscomponentsofthepotentialstriggeredbyavisualoddballtask were averaged over multiple electrode derivations, and the data was analyzed by threewaymultivariateanovaforrepeatedmeasures;severalstatisticallysignificantdifferences werefound.therewerenoafter effectsonbaselineeeg.theinvestigatorsconcludedthat mobile phone EMFs might be beneficial because they inhibited excessive sleepiness. However,thefeweffectsclaimedcouldbeexplainedbychance. Papageorgiou et al. [42] studied the effect of two auditory stimuli on evoked potentials recorded from 15 electrodes, with and without exposure to an unmodulated 900 MHzEMF(no217 Hzpulsemodulation).Threesignificantdifferencesin30testswere found (family wise error rate p=0.19), which did not support the conclusion of the investigatorsthattheemfshadaffectedcognitiveprocessing. Hountalaetal.[43]measuredspectralpowercoherencefrom15electrodesduring an auditorymemory task and concludedthatmales had more coherence than females in the absence of EMFs, 900-MHz EMFs eradicated the differences, and 1800-MHz EMFs reversedit(femalesmorecoherent).theinvestigatorshadpreviouslyreportedthatmales had more spectral power, and that during exposure to 900MHz the power decreased in malesandincreasedinfemales[44].gender baseddifferencesinbrainelectricalactivityin non exposedadultsubjectshadnotpreviouslybeenestablished.theputativedifferences werethereforeconfoundedwiththeputativeeffectsofthemobile phoneemfs. Desynchronization,alsocalledalpha blockingreactions,isachangeinbrainactivity triggered by external events but not time locked to them [6]. Event related desynchronization is normally used to study cortical activation, consciousness, and processing of sensory information preparatory to executing a motor command [6]. In a series of experiments, Finnish investigators evaluated the effects of mobile phone EMF exposure on desynchronization in subjects performing auditory and visual memory tasks [45 49]. The exposure and control sessions typically each lasted 30 min during which approximately200trialswererunandtheeegwasrecordedfrom20electrodes.initially theinvestigatorsdescribedtheoccurrenceofdesynchronizationduetoanemfduringan auditorymemorytask,asassessedusing3 wayanovas[48].however,theresultwasnot replicated in a subsequent study that involved 24 subjects [46]. Desynchronization triggered by a visual memory task was claimed to have occurred in 24 subjects [49]. However, the investigators later revealed that the data had been contaminated by an auditory artifact [45]. They tried again, and claimed to have successfully demonstrated EMF induceddesynchronizationinchildren[45]andadults[47]performingauditoryand visual memory tasks. The effects described were said to depend on details regarding the EMF and on the location of the derivations from which the EEG was recorded. However, bothstudieswerenotprotectedagainstfamily wiseerror. 13
3.2.EffectsDuringExposure:Summary In almost all cases where effects of mobile phone EMFson the EEG were claimed, many statistical tests had been performed without adequate protection against chance. Thus the reports did not provide reliable evidence of an effect of the fields on brain electricalactivity.equallyseriouswastheabsenceofconsiderationoftheroleofelectrode artifacts.ingeneral,metallicelectrodesconnectedtohigh input impedanceamplifierslike thoseusedtomeasuretheeegwillgenerateaspikeattheinputeachtimethestimulusis applied or removed(fig. 3).The duration of the artifact(determined by the pass band of thesystemandtheresponsecharacteristicsoftheamplifier)isapproximately30ms.ina typicalmobile phoneemfexperiment217radio frequencypulsespersecondwereapplied tothesubject,resultingin434electrodespikepotentialspersecond.becausemostofthe studies also applied a low frequency magnetic field in addition to the radio frequency EMFs [7 14, 17 19, 23, 24, 31 41, 45 49], the problem of artifacts was compounded becausethemagneticfieldfromthebatterycurrentalsoproduces434spikes/second.no investigator removed these artifacts prior to decomposing the recorded signal into its componentfrequencies.indeed,onlyoneinvestigatorexplicitlyindicatedthattheartifacts existed[50].inthebaselineeegstudies,thespectralenergyinthespikeswasfoldedinto the true brain signal when the signal was decomposed, and couldhave accounted for the reportedsignificantdifferences. It might be suggested that the problem of spike artifacts did not obfuscate the meaning of the event related studies because the induced potentials were computed by means of time averaging and therefore would have been distinguishable from the spike potentials. However, the induced potentials occurred several hundred milliseconds after the onset of the stimulus, but during the application of the GSM EMF stimulus. Consequently the EEG contained both the temporally locked response to the visual or auditory stimulus as well as the temporally locked spike response due to the GSM EMF stimulus.thustheevent relatedstudieswerealsocontaminatedbyartifacts. Manyreportsemployedlinearanalysis(ANOVA)andconcludedthatmobile phone EMFsincreasedalphaactivityintheEEG[7,11 13,30].Howeversensorystimuliproduce bothincreasesanddecreasesinbrainalpha,dependingonthesubject[51].useofanovato detect stimulus induced changes in that kind of a dependent variable is strongly contraindicated because of the likelihood ANOVA would average away real effects. The thing to be explained, therefore is why so many investigators reported observing what is likelyanon existentphenomenon.ourexplanationisthattheclaimsresultedfromtheuse ofdata miningtomakemeaningfromthedata. 14
Figure 3. Electrode artifacts created when EEG electrodes are exposed to pulsed electromagnetic fields. a) Electroencephalogram (C 3, referenced to linked ears) recordedduringonsetandoffsetof2g,60hzshowingthepresenceoftypicalspike artifacts (~30ms); the artifacts also occur as a consequence of application of electricfields.b)top,electricalnoisefromanelectricalphantomofthehumanhead. Bottom,signalfromthesameelectrodeduringexposuretothemaximumoutputof agsm mobile phone(nokia 6340i); theelectrode was5cmfrom the antenna.the GSM EMF induced artifactual signal consisted of multiple onset and offset spikes produced by radio frequency pulses from the antenna and by low frequency magnetic fields produced by the battery current (217Hz, from each source). The signals in a) and b) were analog filtered at 0.5 35Hz and recorded using an electroencephalograph(nihonkohden,irvine,ca). 3.3. EffectsAfterExposure Vecchio et al. [52] measured the EEG 5 min before and 5 min after exposing 10 subjectsfor45min(onlymales,toavoidpotentialhormonaleffects).duringtheexposure the subjects were able to move around the experimental room and chat with the investigators,whoattemptedtoregulatetheamountoftalking,walking,andthecontentof the conversation so that emotional arousal was not produced. The linear coherence (a measure of the coupling between two signals at a given frequency) among 4 pairs of 15
electrodes was calculated and evaluated using a 3 way ANOVA. An interaction between condition (exposure or sham exposure) and electrode pair was found, which the investigatorsinterpretedtoindicateaneffectoftheemfoninter hemisphericalcoupling of EEG rhythms. However, even assuming the statistical reliability of their interpretation, theuncontrolledactivityofthesubjectscouldplausiblyhaveaccountedforit. Regel et al. [53] exposed 24 subjects for 30 min to either pulse modulated or continuouswaveemfsandcomparedtheeegrecordedpriortoexposurewiththatfrom0, 30, and 60 min after exposure using multiple 3 way ANOVAs. They claimed increased poweronlyat10.5 11Hz,onlyat30min,onlyforthepulsedfield,andonlyafterignoring thepotentialroleofchance. Perentosetal.[50]studied12subjectswhowereexposedfor15min.TheEEG(16 channels) was analyzed immediately after EMF exposure because spike artifacts were present in the EEG signal measured during exposure. No change in EEG power in any frequencybandwasseen. Severalreportsinvolvedtheeffectsofmobile phoneemfsontheeegduringsleep [54 58].Huberetal.[54]exposed16subjectsfor30minduringanawakeperiodpriortoa 3 h sleep episode and found an increased power in some of the alpha frequencies during thefirst30minofnon REMsleep.Loughranetal.[55]repeatedthestudyandalsofound increased alpha power from the same two derivations (C3 and C4) (although not at the same frequencies). Taken together the reports suggested a time dependent after effect. However, neither study was adequately protected against family wise error because 3 significantdifferencescouldhaveoccurredbychancein24tests(8 14Hz,0.25Hzbins)in two independent studies). In a similar experiment Huber et al. found an after effect on alpha power in stage two sleep [56]. Regel et al. [57] repeated the study and found an increaseinsomealphafrequenciesinnon REMsleep,butnotinstagetwosleep. Hung et al. [58] exposed subjects for 30 min and reported increased delta power duringthesubsequent90min,butonlyin1of6derivations. Threereportsfailedtofindaneffectofmobile phoneemfsonauditorybrain stem evokedpotentials;stefanicsetal.[59]andparazzinietal.[60]after10minexposure,and Araietal.[61]after30minexposure. Inaseriesofexperiments,Japaneseinvestigatorsaskedwhether30minexposureto GSMmobilephones(800MHz,0.8maximumoutputpower)alteredpost exposurehuman brain activity[62 64]. The exposure did not affect somatosensory evoked potentials[62], the frequency of visual saccades [63], or motor evoked potentials elicited by pulse transcranial magnetic stimulation (a result that was not surprising even to the investigators)[64]. Magnetoencephalography involves the recording of weak magnetic fields outside the head; it is a far rarer method than the EEG for characterizing brain electrical activity butmayhavesometheoreticaladvantagesforunderstandingbrainfunction[6].hinrichset al.[65]employedmagnetoencephalographytostudytheeffectsofmobile phoneemfson 16
memoryencoding,usinganencoding retrievalparadigm.thesubjectswereexposedtothe field while they memorized a list of words and then were asked to recall specific words. The EMF exposure altered the amplitude of the event related magnetic field, but only withinanarrowlatencyrangethatwasidentifiedbydata mining. 3.4. EffectsAfterExposure:Summary The advantage of after effects experiments was that they obviated the problem of electrodeartifacts.thedisadvantage(inadditiontothosediscussedinconnectionwiththe during exposure experiments) was that the after effect design undercut the rationale for studying EMF effects on brain electrical activity. In the reports discussed previously the EMFs were physically present in the brain during all immediate early biological events including putative signal transduction, generation and propagation of an afferent signal, andsignalprocessingatvariouslevelsofbrainorganization.atleastinprinciple,therefore, the studies were pertinent to the hypothesis that mobile phone EMFs affected cognitive processing.theafter effectsstudies,incontrast,werebioassaysatbestbecausetherewas simply no way of ever knowing whether any observed changes were simply general metaboliceffectsorspecificeffectsonthebrain.bioassayexperimentswereappropriateat the dawn of the EMF health hazards issue [66], but they no longer serve any important purpose because it is now universally agreed that EMFs can cause biological effects further affirmative answers to the can question are unneeded. The present task is to conducthypothesis drivenstudiesthathelpprovideadeeperunderstandingofparticular effects. 4. Discussion 4.1.AnalysisofReports A fundamental and characteristic scientific limitation in the reports was that the experiments were based on an incorrect electrophysiological model of the brain, that its electrical activity was in equilibrium with its surroundings. For a given subject, the instantaneous values of the dependent variables(spectral power, evoked potentials, slow potentials, as examples) measured and/or computed from various scalp derivations were assumed to be stationary, and therefore representable by a time average value. Intersubject differences were conceptualized as a Boltzmann distribution that could be representedbyagrandaverage.inthisperspective,oneinwhichdynamicalactivityhadno physiologicalsignificance,theeffectofmobile phoneemfsonthebraincouldbe(andwas) assessedusinganovas(orotherlinearstatistics)todeterminewhetherthegrandaverage differedbetweenthepresenceandabsenceofthefield,orbetweenpre andpost exposure. This model was adopted in all but four reports [23, 24, 26, 27], even though it was antitheticaltotheacceptedviewofbrainfunction[67 75].Thebrainisnotinequilibrium withitsenvironment,theeegisnotstationary,andtheeffectsonbrainelectricalactivity arenotlinearlyrelatedtoemfs(seebelow).fromthebeginning,therefore,thepossibility 17
thattheexperimentscouldprovidereliableinformationwasnilbecausetheirfoundational assumptionwasinconsistentwithhowthebrainworks. We showed that magnetic fields comparable in frequency and strength to those produced by the battery current in GSM mobile phones consistently caused changes in brain electrical activity in human subjects, and that the changes could be detected only when the EEGs were analyzed using nonlinear methods [1, 8, 76 80]. Whether brain electrical activity is also affected by radio frequency GSM EMFs is an empirical question. Based on results involving older mobile phone technology [81], a model of a candidate transductionmechanism[82],andempiricalobservationsthatthefrequencyofanemfis not a primary factor in determining the resulting biological effects [66], we think that properly designed and performed studies likely will reveal that mobile phone radiofrequencyemfsconsistentlyaffecthumanbrainelectricalactivity.ifanunmodulatedhighfrequency mobile phone field is applied via an antenna, an onset evoked potential with a latency of 100 400ms will occur, depending on the subject. If the field is maintained beyond the decay of the onset potential, an effect on brain activity due to the presence of theemfwillbeobserved,andwillpersistforthedurationofthestimulus[76,79].when the mobile phone field is terminated, an offset evoked potential occurring 100 400ms later will also be detected [76]. If the field is applied as a brief pulse, say 50ms, as is commonly the case in evoked potential studies, then field induced evoked potentials consistingofsuperimposedonsetandoffseteffectswillbeobservedwithalatencyof100 500ms,dependingonthesubject[78].Nonlinearanalyticaltechniquesareaprerequisite for observing reproducible effects of GSM EMFs on human brain electrical activity; recurrencequantificationanalysisisonesuchtechnique[1,8,76 80]. Even if nonlinear analysis were employed, it would not be a substitute for a reasonable experimental design both elements are needed. For example, when a highfrequency mobile phone EMF is applied in a manner that simulates normal phone use (multiple bursts of high frequency energy over a period on the order of minutes), a nonlinearcombinationofnumerousrepetitionsoftheelementalprocessdescribedcanbe expected. The experimental design must therefore accommodate this complexity. If the high frequencyemfisappliedbymeansofamobilephone,additionalcomplexeffectsdue to the battery current magnetic field can be anticipated. If the EMF exposure takes place while the subject is cognitively processing auditory or visual information, still further complexity is unavoidable. Use of idiosyncratic experimental designs [36 38, 45 49] or noveldependentvariables[43,44]arealsocomplicatingfactors. The consistent absence of due regard for family wise error in evaluating the meaning of the statistical tests was another serious problem. When many statistical tests areperformedtoexaminethehypothesisthatmobile phoneemfsalteredbrainelectrical activity,andonetestissignificantatp<0.05,rejectionofthenullhypothesisisnotjustified at a confidence level of PFW<0.05. A variety of verbal formulas were used to express a conclusion that mobile phone EMFs had altered brain electrical activity. As examples, 18
investigators asserted that EMFs caused clear tendencies [30], induced but not evokedbrainpotentialactivity [35], asignificantdecreaseofslowpotentials [37], improve(d)performance [41], agender relatedinfluenceonbrainactivity [44], altered (desynchronization) responses [49], effects on brain oscillatory responses [45],a (modified) sleepelectroencephalogram [55], adose dependent increase of power [57]. The formulas were misleading because they obscured the fact thattheassociatedlevelofchancewasgreaterthan5%. AthirdprobleminvolvedtheexperimentswheretheEEGwasmeasuredduringfield exposuretomobile phoneemfs.wheneverasubjectwithscalpelectrodesisexposedtoan EMF, the possibility must at least be considered that the onset and offset of the field triggeredaspikeartifactattheinputoftheeegamplifier(fig.3).manyfactorsincluding thepowerdensityoftheemf,thespatialrelationbetweentheelectrodesandtheantenna, and the characteristics of the measuring circuitry will affect the detectability of the spike artifacts. The point is not that they will always occur, but rather that their possible presencemustbeassessedineachstudy.theartifactsmaybeidentifiedandremovedprior to data analysis, excluded as a likely source of error on the basis of proper argument, or acknowledged as a valid alternative explanation to a claimed EMF induced physiological effect.toignoretheartifactissue,however,isimpermissible.yettheissuewasignoredin almostallthereports,andwasnotadequatelyaddressedinevenonecase.thepresenceof the spikes constituted an alternative explanation for eachof the claims of positive effects that were based on statistically significant differences between EEGs compared in the presenceandabsenceofgsmemfs. Fourth, with few exceptions, the investigatorsdid not employ positive or negative controls. Thus, where an effect was claimed, the absence of sham experiments (shamexposure compared with its corresponding control) left open the possibility that the claimedeffectwasattributabletotheanalyticaltechniquesused,whichinsomecaseswere so complex as to beggar description[31 33].The absence of positive controls essentially eliminated any potential value in the 18 negative studies (Table 1) because, for all we know,nonon lethalstimuluscouldhaveproducedsignificantdifferencesinthedependent variables. This considerationis particularly applicable to the brain stem auditory evoked potential studies [34, 59 61], which were all negative and could hardly have been otherwise. 19
Table1.ReportsinvolvingtheeffectsofGSMelectromagneticfieldsonhumanbrain electricalactivity.n=55.f,foundation;r,research.+( ),studyself reportedas positive(negative). REF NO. [CLAIM] SUPPORT 42[+] AstraZeneca Hellas 11[+] Clarus Products 41[+] COST281 (F. Funk) 34[ ] EARnEARE (Nokia) 26[+], 27[+],28[+], 29[+], 30[+] Estonian Science F. 53[+], 54[+], 56[+] Foundation for R. 35[+] German R. F. 7[+],9[ ],17[+], 18[ ], 19[ ], 23[+], 24[+], 36[+], 37[+] German Telekom 43[+], 44[+] Greek Ministry 59[ ], 60[ ] Guard (Nokia) 52[+] Italian Telecom 65[+] KPN Mobile 57[+] Mobile Communication F. 31[+], 32[+], 33[+] Mobile Communications Consortium 58[+] Mobile Phone R. 13[+] Motorola 45[+], 46[ ], 48[+], 49[+] Nokia 22[ ] Radio R. 50[ ] Radiofrequency R. Center 61[ ], 62[ ], 63[ ], 64[ ] Radio-industry & business committee 47[ ] R. Association for Radio Applications 55[+] R. Council (Nokia) 15[ ],16[ ],20[+], 21[+] Swisscom 10[ ] Tech. Dev. Center 12[+], 14[+], 39[+], 40[ ] Telstra R. Laboratories 20